Dopamine is a neurotransmitter, a chemical that shapes how the brain processes information. It does this by binding to different categories of dopamine receptors which then leads to changes in the intracellular processes of neurons, the cells responsible for transmitting information in and outside the brain. The D1 family of dopamine receptors (D1 & D5) increase intracellular levels of a chemical second messenger, cyclic AMP, which can then affect how a neuron processes other signals it receives. The D2 family of dopamine receptors (D2, D3, & D4) decrease cyclic AMP, which can also shape neural responses. How the dopamine signals interact with other signals in the brain can be quite complex and is beyond the scope of this piece. For more see this review article.

Dopamine signaling plays a role in a variety of critical cognitive processes including motor control, learning, and decision making. It has also been implicated in the addictive nature of drugs of abuse, which I studied in some detail during my Ph.D. and postdoctoral research. 

Positron Emission Tomography (PET) allows scientists to measure dopamine signaling in the living brain. PET has been around since the 1960s and involves imaging the location and amount of a radiotracer (radioactively-tagged compound) in the body. Most PET radiotracers contain C-11, F-18, or O-15 radioactive isotopes. These isotopes release positrons (which are the antiparticle of the electron) which, when they interact with nearby electrons in the body produce an annihilation event leading to 2 gamma ray photons being emitted at 180 degrees. The PET scanner "counts" these gamma ray events and ultimately reconstructs the image that produced the events by projecting the gamma ray counts back into the body part being imaged. These PET images give quantifiable data regarding the amount of tracer that accumulates in a particular area over time.

​Brain PET is a particularly powerful technique in that we can use radiotracers that allow us to investigate brain metabolism, neurotransmitter receptors (dopamine or opioid, among others), neurotransmitter synthesis, and the presence of beta-amyloid plaques (often present in Alzheimer's disease). With these compounds we gain a better understanding of individual differences that may be useful as markers of disease state or risk for developing a particular disease. Common radiotracers for imaging the dopamine system include FDOPA, C-11-Raclopride, F-18-Fallypride, FMT, and others. Several groups have used some of these compounds to better understand the dopamine system's role in drug abuse. 

There has also been interest in understanding whether genetic differences may lead to different levels of D2 receptor availability, potentially placing some individuals at greater risk for addictive disorders. I investigated the effect of some common D2 receptor single nucleotide polymorphisms (SNPs) on D2 receptor availability using F-18-Fallypride as part of my postdoctoral research. Many of these SNPs had been previously associated with dopamine receptor differences in relatively small PET studies or been associated with potential increased risk for drug addiction. 

• Taq1A - A1 allele associated with lower striatal D2 receptor availability (replicated in separate study but not in a third)
• C957T - C allele associated with lower striatal D2 receptor availability in study of 45 individuals 
• -141C Ins/Del - inconsistent findings on whether it affects D2 receptor availability

For more see: Genetic variation and dopamine D2 receptor availability: a systematic review and meta-analysis of human in vivo molecular imaging studies

Since the Taq1A SNP was discovered to associate with differences in dopamine signaling first, researchers have used it as a proxy for D2 receptor status (or more loosely as an index of general dopamine functioning). However, given that the Taq1A polymorphism does not occur within the DRD2 gene itself, researchers have speculated that polymorphisms in Taq1A may associate with other SNPs in the DRD2 gene that are the real drivers of expression of the receptor in vivo.

The C957T and -141C Ins/Del polymorphisms are in strong linkage disequilibrium with Taq1A and have themselves been associated with striatal D2/3 receptor availability. Despite the data suggesting that these SNPs are strongly linked, few studies have systematically investigated the effect of C957T, -141C Ins/Del, and Taq1A in isolation and combination on D2/3 receptor availability. Beyond the potential link to drug addiction risk, characterizing the functional effect of these SNPs on D2/3 receptor availability has implications for better understanding the mechanisms through which they exert their demonstrated influence on motivated behaviors including learning and decision making, impulsivity, and reward responsivity. 

In our work, we used F-18-Fallypride, which is a D2/3 receptor tracer with favorable affinity to measure both striatal and extrastriatal dopamine receptors, and assessed the impact of C957T, Taq1A and -141C Ins/Del SNPs on D2/3 receptor availability in a sample of 84 healthy subjects.

We found that the C957T SNP was associated with variation in dopamine D2/3 receptor availability in areas of the striatum often implicated in reward processing. The fact that the C allele was associated with lower dopamine receptor availability suggests it could be a useful genetic measure for at least one biological factor (lower D2 receptor availability) linked with drug addiction. While more work needs to be done to confirm these results, certainly further study of the C957T SNP in the DRD2 gene is warranted. 

Using this PET technique, Casey et al 2014 found that young adults with a multigenerational FH of substance use disorders showed reduced dAMPH-induced dopamine than either healthy controls or subjects that personally used drugs at similar levels to the FH group but without a FH of substance use disorders. This study was particularly informative as the effects of current drug use were also investigated and measured separately from family history. Furthermore, our group and others have demonstrated that dAMPH-induced dopamine release correlates with subjective ratings of the drug, particularly wanting more, in drug naïve individuals. These data confirm animal work linking changes in dopamine signaling after drug use to wanting processes (which has been labeled incentive salience).

Read more about wanting, liking, and drug abuse in a previous blog post.
​
The concept of blunted dopamine signaling (lower D2 receptor levels and less dopamine release) as biomarkers of addiction has also been recently reviewed (Trifilieff et al 2017; Leyton, 2017). While more work needs to be done, understanding factors that influence these PET-based biomarkers of dopamine signaling in human subjects has the potential to identify at risk individuals. This risk identification may allow intervention to be attempted earlier in the addiction process or perhaps prevent addiction before it even occurs.

The rapid reduction in the cost to sequence the human genome (complete set of an individual’s DNA) as well as proliferation of genotyping services such as 23andMe (which genotype common genetic polymorphisms, or areas in human DNA most likely to vary across individuals) means that genetic data can be readily obtained by anyone who wants it. This technological advance will allow physicians greater information of a patient’s underlying biology and eventually will be merged with growing insights into the effects of genetic variation on drug metabolism, brain signaling, and behavior to make personalized medicine commonplace. In fact, pharmacogenomic data has been added to several drugs by the FDA.

My own work, referenced above, suggests that genetic variation in a gene encoding the dopamine D2 receptor (DRD2) can affect the relative availability of this receptor in the brain as measured with PET (Smith et al., 2017 Translational Psychiatry). Individuals with a particular genetic variant in DRD2 that is associated with less availability of the receptor (C957T CC individuals) may need either a higher dose of a D2 drug or a higher affinity D2 drug to receive a therapeutic benefit.

The implications for this finding go beyond potential treatments or interventions for drug addiction. D2 agonists are commonly used in Parkinson’s Disease patients to preserve motor function and D2 antagonist-like drugs are used in the treatment of Schizophrenia. Understanding the genotype of individuals affected with these conditions, then, could enhance the effectiveness of their D2 drug treatments (by suggesting a physician might want to start with a higher or lower dose of the drug). While studies such as ours linking genetic variation with differences in biology are encouraging, DNA can also be modified by the environment. Researchers have begun studying these epigenetic effects on behavior, with most work occurring in rodents. As we integrate this knowledge, we will begin to better understand the impact gene by environment interactions have on biology and behavior.

Genetics are not the only variables that could be worth attending to in future treatments. Additionally, dopamine signaling is known to decline with age (see also a previous blog post on this topic). So, doses of dopaminergic drugs that work well on young adults might need to be titrated in older adults. Furthermore, we and others have shown that estradiol levels in naturally cycling women can affect dopaminergic brain functions (assessed by fMRI imaging and a genetic variant (COMT) know to affect dopamine levels in the higher-order, prefrontal areas of the brain). Thus, a dopaminergic medication might be more effective at treating a female patient’s symptoms at certain points of her menstrual cycle but not others. We are only beginning to understand the role of female sex hormones in a variety of biological systems as basic research historically has focused on male model organisms.​

The role of dopamine in drug addiction is quite complex. In addition, implementing personalized medicine when treating psychiatric or behavioral disorders is challenging as most of these disorders do not have a single, identifiable biological cause. The brain is complex enough and the fact that genetics, sex hormones, age, and environment can all affect one neurotransmitter (dopamine) among the many others involved in brain function speaks to the vast challenge that lies ahead for researchers.

​Our quest to better understand individual differences, however, has the potential to lead to more targeted treatments and therapies for a variety of dopamine-associated disorders including ADHD, Schizophrenia, Parkinson’s Disease, and drug addiction. The development of these personalized treatments will undoubtedly improve healthcare in the 21st Century and beyond but will require further research focused on measuring and categorizing individual differences. ​

It has been over 7 years since I defended by Ph.D. dissertation in March 2014 at the University of North Carolina at Chapel Hill. Here, I wanted to share some of the rationale and implications of my graduate research on immediate reward selection bias in humans. While this research encompassed 5+ years of my life and resulted in a 112-page dissertation, I will focus on the key points and findings and why they are important. I have moved on from doing this research in my current role but it will forever be a part of my identity. In addition, I hope my scientific contributions have added a bit more to our understanding of substance abuse risk factors and how we might work to either intervene to support those at risk for addiction proactively or treat some of the behavioral patterns in addicted individuals that can continue the cycle of problematic drug use despite its negative consequences. 

For an intermediate phenotype to be useful it must be a quantitative, continuously variable feature or behavior that can be consistently measured. Furthermore, as these intermediate phenotypes are thought to convey genetic risk for a disorder, they should be elevated in those affected with the disorder as well as in those individuals’ close relatives who share genetic similarity with them. Importantly, the level of these phenotypes in affected individuals and their close relatives should be shifted away from a distribution of those otherwise unaffected with no familial risk (Gottesman and Gould, 2003). For example, Egan et al. (2001) found unaffected siblings of those with schizophrenia to display executive function deficits that fell between unaffected nonrelatives and individuals with schizophrenia.

A variety of criteria have come to define an intermediate phenotype in psychiatry (Almasy and Blangero, 2001; Gottesman and Gould, 2003; Waldman, 2005; Meyer-Lindenberg and Weinberger, 2006):

1. The phenotype should be sufficiently heritable with genetics explaining variance in the behavior.
2. The phenotype should have good psychometric properties as it must be reliably measurable to be a useful diagnostic.
3. The phenotype needs to be related to the disorder and its symptoms in the general population.
4. The phenotype should be stable over time in that it can be measured consistently with repeated testing, potentially to assess treatment effects.
5. The behavior should show increased expression in unaffected relatives of those with the disorder as highlighted by Egan et al. (2001), above.
6. The phenotype should co-segregate with the disorder in families in that a family member with the disorder should show the behavior or trait to a greater degree than an unaffected sibling and that this unaffected sibling should display the trait to a greater degree than a distant unaffected relative.
7. The phenotype should have common genetic influences with the disorder.

To illustrate the intermediate phenotype concept and associated criteria, we can look at research in schizophrenia. Schizophrenia is associated with poor performance (and hyperactivity in an area of the brain known as the dorsolateral prefrontal cortex, dlPFC) on executive function tasks. As mentioned above, Egan et al. (2001) found unaffected siblings of those with schizophrenia to display executive function deficits that fell between unaffected nonrelatives and individuals with schizophrenia. Furthermore, genes affecting dlPFC activity and executive functions such as the catechol-O-methyltransferase (COMT) gene explain variation in schizophrenia risk (see Egan et al., 2001). Thus, by investigating a specific behavior (executive function) and its neural correlate (dlPFC activity) in schizophrenic patients and those at increased genetic risk for the disorder, a genetic factor (COMT) was isolated. Schizophrenia is caused by more than one genetic variation but this example illustrates the value of identifying a link between a behavior associated with schizophrenia (an intermediate phenotype) and a potential biological and genetic basis for said behavior.  ​

Delay discounting (DD) behavior reflects the tendency for animals (including humans) to discount the value of delayed (in time) rewards in comparison to those available immediately. DD has also been referred to as immediate reward selection (“Now”) bias as the value of rewards available immediately supersedes waiting for a larger, delayed reward in the future (Rachlin and Green, 1972). Other terms for this behavior include temporal discounting or hyperbolic discounting as plots of the value of time-delayed ​rewards relative to present value often take a hyperbolic shape with the present value of a reward delivered at longer delays decreasing at a steep, non-linear rate. In other words, $100 in 3 months may only be worth $25 in present value while $100 in 1 month is worth $50 in present value (the discount rate gets steeper moving from a 1-month to a 3-month delay). 

As delay discounting (DD) behavior has been shown to be highly heritable (Anokhin et al., 2011; Anokhin et al., 2015; Mitchell, 2011), suggesting a strong genetic component, and is elevated in a variety of addictive behaviors (MacKillop et al., 2011), we focused our current work of exploring intermediate phenotypes for addiction on this behavior. Prior work has suggested DD displays many of the necessary criteria of an intermediate phenotype for a variety of neurobehavioral disorders including substance use disorders (SUDs) (Becker and Murphy, 1988; Reynolds, 2006; Perry and Carroll, 2008; Rogers et al., 2010), attention deficit hyperactivity disorder (Barkley et al., 2001; Sonuga-Barke et al., 2008; Paloyelis et al., 2010), and pathological gambling (Alessi and Petry, 2003; Leeman and Potenza, 2012). As these behaviors often co-occur, they may share similar biological and genetic components (Wilens, 2007; Leeman and Potenza, 2012).

As Now bias is elevated in those with AUDs, we might expect to see this behavior heightened in those on a trajectory toward an AUD as well. Such demonstrations between elevated Now bias and AUD risk would add greatly to the utility of Now bias as an intermediate phenotype. As problematic alcohol use during emerging adulthood (late teens to early twenties) may predict development of an AUD later in life (O'Neill et al., 2001; Merline et al., 2008; Dick et al., 2011), though many individuals mature out of problematic use (Bartholow et al., 2003; Costanzo et al., 2007; Lee et al., 2013), one might expect Now bias is enriched in problematic drinking emerging adults. Only one relatively small behavior study has looked at such a relationship with Now bias observed to be heightened among heavy versus lighter social drinking college students (Vuchinich and Simpson, 1998). This finding requires replication in a larger, more diverse sample.

In addition to being elevated in problematic drinking emerging adults, to satisfy another intermediate phenotype criterion for AUDs, Now bias behavior should also be elevated in unaffected first-degree relatives (parents, siblings) of those suffering from AUDs (intermediate phenotype criterion 5). Elevated Now bias in first-degree relatives of those with AUDs has yet to be adequately demonstrated, however.

​Most of the intermediate phenotype literature considers the expression of the behavior or trait in first-degree relatives as critical in demonstrating that behavior as an intermediate phenotype. In the field of AUDs, however, positive family history of an AUD is often defined as having at least one parent with an AUD (Acheson et al., 2011) or father with an AUD (Crean et al., 2002; Petry et al., 2002), or some combination of parental history or sufficient density of AUD history in second-degree relatives (Herting et al., 2010). In these previous studies, the effects of family history on Now bias was either only observed in females (Petry et al., 2002), was not found at all (Crean et al., 2002; Herting et al., 2010), or was not present when controlling for group differences in IQ and antisocial behavior (Acheson et al., 2011).

Measuring Now bias behavior in individuals with any first-degree relatives with AUDs expands the classic family history positive AUD definition to include siblings, who display greater genetic concordance with a particular individual than their parents. To our knowledge, though, this definition of first-degree family member positive or negative for AUDs has not been applied to the study of Now bias. Thus, while Now bias possesses many properties that suggest it could be a good intermediate phenotype for AUDs, further investigation of this possibility is warranted, particularly work focusing on examining whether Now bias is elevated in unaffected individuals with first degree relatives with AUDs.

Given our review of the literature and past work in this area (Mitchell et al., 2005), we focused on better demonstrating the utility of Now bias as an intermediate phenotype for AUDs in a large group of individuals who ranged in age, level of alcohol use, and family history of AUDs. This work was published in Frontiers in Human Neuroscience in 2015, but I share the key takeaways from the study, below. ​

As the Alcohol Use Disorders Identification Test (AUDIT) is an effective means of measuring problem drinking behavior (Fiellin et al., 2000; Barbor and Higgins-Biddle, 2001; Kokotailo et al., 2004), we recruited high and low AUDIT individuals across a group of 18-40 year old social drinkers not reporting any AUD. We hypothesized that Now bias would be elevated in high but not low AUDIT emerging adults (defined as ages 18-21 or 18-24). Furthermore, we sought to test whether Now bias was elevated in those otherwise unaffected individuals (light/moderate social drinkers; low AUDIT) with a first degree relative with an AUD. We used the intermediate phenotype criteria of first-degree biological relative status (father, mother, or sibling with AUD), excluding those with mothers with an AUD to rule out potential fetal alcohol effects. We hypothesized that Now bias would be elevated in low AUDIT individuals with a first-degree relative with an AUD but not in those with no first-degree AUD relative. ​

As this study was underway, we began to wonder how age might impact Now bias independent of problematic alcohol use. Our lab had previously found marked Now bias among emerging adults (18-25 yrs), regardless of drinking behavior. This suggests elevated DD generally among individuals transitioning from adolescence to adulthood. The observation that adult controls (average age of 26-28) with no AUD diagnosis display reduced Now bias compared to abstinent alcoholic adults (Mitchell et al., 2005; Boettiger et al., 2007) suggests that this bias should decline between emerging adulthood and adulthood, at least among moderate, non-problem drinkers. While emerging adults are widely regarded as impulsive (Chambers and Potenza, 2003; de Wit, 2009), and DD normally decreases from childhood to the early 30’s (Green, 1994; Scheres et al., 2006; Olson et al., 2007; Eppinger et al., 2012), little is known about specific changes in DD from late adolescence to adulthood. Some data show trait impulsivity declining linearly with age from early adolescence to age 30 (Steinberg et al., 2008). Thus, given positive correlations between DD and trait impulsivity (Mitchell et al., 2005; de Wit et al., 2007), we hypothesized DD should decline with age from adolescence into the 30s, but, to our knowledge, no prior studies have explicitly investigated age effects on DD in detail from ages 18 to 40. Moreover, we do not know whether heavy alcohol use moderates any such age-related changes in DD. Thus, a secondary aim of our work was to investigate age-related differences in Now bias in our population as a whole and separately in those reporting heavy, problematic versus light/moderate drinking. 

Confirming and extending on prior work, we found that emerging adults (defined as either aged 18-21 or aged 18-24) regardless of their drinking status (light/moderate vs heavy drinkers) showed equally high Now bias behavior, which did not support our first hypothesis that this behavior would be elevated in heavy drinkers. Follow-up analyses concluded that Now bias generally declined with age in our light/moderate drinker population (r=-0.28, p=0.022) but not in the heavy drinkers (r=-0.03, p=0.39). We measured Now bias in our study as an impulsive choice ratio (ICR), which can range from 0 (no Now bias) to 1 (complete Now bias). The age-related decline in Now bias began to asymptote around age 25. Thus, we organized our data into emerging adult (aged 18-24) and adult (aged 26-40) groups for further analyses. 

Our work has added additional support to Now bias being an intermediate phenotype for alcohol use disorders (AUDs). The fact that Now bias was elevated in heavy drinking adults without an AUD was suggestive that this behavior may proceed and AUD diagnosis. More work is needed to follow up on this finding, however. Specifically, longitudinal studies need to be conducted to measure Now bias in individuals in their early teens (prior to exposure to drinking) and continue to measure this behavior over the lifespan, especially as these individuals enter their late teens and early twenties when problematic drinking behavior often emerges. Only through careful study of the trajectory of Now bias during adult development in both non-problematic and problematic drinkers can we begin to truly determine the utility of this measure as an intermediate phenotype for alcohol use disorders or substance use disorders in general.

Ongoing work taking place as part of the Adolescent Brian Cognitive Development (ABCD) Study seeks to understand adolescent brain and cognitive development generally and the various behavioral (including Now bias) and neural risk factors that can emerge in adolescence that lead to mental or psychiatric disorders in adulthood.

Learn more about this ambitious study here and view current publications emerging from the dataset here. 

Since our study on Now bias as a potential intermediate phenotype for AUDs was published in Frontiers in Human Neuroscience, other work has shown:

• Large individual differences in intertemporal choice behavior (Review; Keidel et al., 2021)
• Genomic basis of delayed reward discounting (Gray et al., 2019)
• Steep Discounting of Future Rewards as an Impulsivity Phenotype: A Concise Review (Levitt et al., 2020) 
• ​Individuals with two parents with addiction have significantly higher rates of discounting compared to those with no or only one parent with addiction (Athamneh et al., 2017)
• The density of familial alcoholism interacted with binge-drinking status to predict impulsive choice (Jones et al., 2017)
• A review of age & impulsive behavior in drug addiction (Argyriou et al., 2017)

With increased confidence in Now bias as an intermediate phenotype for alcohol use disorders, our next step is better understanding the neural and biological bases of this behavior. This information may then offer a means to potentially reduce Now bias in individuals at risk for alcohol use disorders. Making at-risk individuals more future-focused could assist them in considering the long-term consequences of problematic alcohol use and reduce the temptation to drink heavily in the moment. Targeting the dopaminergic system is one potential approach to modulating Now bias as some of my and others' work has shown. Delving into that topic will have to wait for a future post, though. Stay tuned. 

Explore more of my work on Now bias:

• Age modulates the effect of COMT genotype on delay discounting behavior
• Ovarian cycle effects on immediate reward selection bias in humans: a role for estradiol 
• Modulation of impulsivity and reward sensitivity in intertemporal choice by striatal and midbrain dopamine synthesis in healthy adults
• Neural Systems Underlying Individual Differences in Intertemporal Decision-making  

And additional neuroscience topics on the blog:

• Declining Dopamine: How aging affects a key modulator of reward processing and decision making
• Stress and the Brain: How genetics affects whether you are more likely to wilt under pressure
• Wanting, Liking, & Dopamine's Role in Addiction 

I have written about weighing the value of pursuing a postdoc after completing your Ph.D. in a previous blog post. And while that post focuses on my own personal experience and opinion, I wanted to use this space to emphasize more practical advice on how to network your way to postdoc opportunities and consider the training environment in the lab/group/institution to make the most of your postdoctoral experience.

I am often asked by Ph.D. students how they should search for postdoc positions and make the most of them. Recently, I gave a presentation on the topic to a graduate class at North Carolina State University. I am sharing it here for those of you debating should you do a postdoc, how to find a good postdoc environment, and how to make the most of this training period.

The United State's National Science Foundation (NSF) collects a wealth of data on individuals who received their doctorate degrees from US universities. Back in April, they released their most recent batch of data from their 2019 Survey of Doctorate Recipients (SDR). The SDR provides demographic, education, and career history information from all individuals with a research doctoral degree in a science, engineering, or health (SEH) field from a university in the United States. As the SDR seeks to capture the full scope of US SEH Ph.D. employment, it surveys anyone with a Ph.D. in SEH fields from a US university regardless of year of graduation: some SDR respondents received their Ph.D.s a few years ago and some 20+  years ago. This is different from the Survey of Earned Doctorates (SED) which surveys new US Ph.D. recipients and whose data I shared in an earlier blog post. Here, I will delve into some of the trends observed in the 2019 SDR data to give those with a Ph.D. in a SEH field more insights into employment possibilities after they receive their degree.  

Across all doctorate recipients surveyed in the 2019 SDR, the US states with the largest proportion of science, engineering, or health Ph.D.s employed in them included the District of Columbia (technically not a state but represented in the state-level data; ~2.5% of the population are SEH Ph.D.s), Massachusetts (0.8%), and Maryland (0.6%). The median percent of any state's population consisting of employed SEH Ph.D.s was 0.2%. While DC, Massachusetts, and Maryland remained the top states employing biological, agricultural, and environmental life science Ph.D.s, others with high proportions of bio science Ph.D. employment included Vermont, Montana, Connecticut, North Carolina, Nebraska, California, & Washington state. Note that as these are calculated as proportion of a state's 2019 population, states with relatively low population counts (Vermont, Montana, & Nebraska) have many less Ph.D. scientists employed in them than larger states. For example, according to the 2019 SDR, there are 32,900 biological, agricultural, and environmental life science Ph.D.s employed in California (with a population of 39.5 million in 2019) while Vermont has 600 (among a population of ~600,000). 

Top 10 states for employing computer science Ph.D.s: DC, Washington, Massachusetts, California, Maryland, New York, Utah, Virginia, Oregon, New Jersey

Top 10 states for employing physical science Ph.D.s: DC, Delaware, Massachusetts, Maryland, New Mexico, Colorado, Oregon, Connecticut, California, New Jersey 
And many of these states are also top employers of engineering Ph.D.s.

Given these data you may have more luck pursuing Ph.D.-level employment in certain areas of the country over others. 

Across all SEH Ph.D.s surveyed in 2019, ~3.3% of those employed worked as a postdoc. However, the percent of employment represented by postdocs varied by field of doctorate with ~6.1% of biological, agricultural, and environmental life science Ph.D.s employed as postdocs while ~1% of computer science Ph.D.s were employed as postdocs. The percentage of engineering Ph.D.s employed as postdocs was ~2%.

Given a postdoctoral position is by definition temporary, one would expect the percent of all employed SEH Ph.D.s in a postdoc would be rather low. While the general proportion of Ph.D.s employed in postdocs is relatively low, some of the trends in postdoctoral employment are concerning. 

​Unfortunately, many postdocs have been in their positions longer than the 5 year post-Ph.D. guidance outlined by The National Academies of Sciences, Engineering, and Medicine's The Postdoctoral Experience Revisited report released in 2014 (see press release). According to the 2019 SDR data, 19% of all science postdocs were >5 years from the date of their Ph.D. being awarded and this percentage was slightly higher (21.3%) for biological, agricultural, and environmental life sciences postdocs. So, as many as 1 in 5 postdocs employed in the US are 5+ years past receiving their terminal degree.

In addition, over the past 10 years a larger proportion of the US postdoctoral population is being filled by those 5+ years post-Ph.D. In the 2010 SDR data, only 13.1% of all science postdocs and 14.9% of bio, ag, and environ life science postdocs were >5 years from their Ph.D. being awarded. And while the 2019 data is off the peak of >25% of postdocs >5 years from their Ph.D. seen in 2015, the proportion of Ph.D.s employed as postdocs >5 years from their terminal degree is still ~45% higher in 2019 than 2010. 

So, while improvements have been made around limiting long postdoctoral training periods, more needs to be done to assist these individuals in transitioning into more permanent positions either within or outside academia. 

Across all science ("all science" refers to all SEH fields surveyed except engineering and health) Ph.D.s surveyed in the 2019 SDR, ~48% work for and educational institution while ~30% are employed by a for-profit company and ~8% work for either the federal or state government. The distribution of sectors employing Ph.D.s in 2019 differed markedly by the field of the individual's doctorate degree with employment by educational institutions quite high for the social sciences (~67% of employed Ph.D.s) and for-profit companies being the largest sectors employing computer & information science (~54%) and engineering Ph.D.s (~58%).

While educational institutions are the top employers of social science Ph.D.s, they employ ~47% of those with Ph.D.s in the biological, agricultural, and environmental sciences. The proportion of engineering and chemistry Ph.D.s employed by educational institutions is even less with for-profit companies employing 50%+ of Ph.D.s from these fields. 

These data suggest certain sectors of employment may be more available to particular Ph.D. fields than others. It is difficult, however, to disentangle whether engineering and chemistry Ph.D. skills, for example, are more valued by for-profit companies than those in the social sciences or whether there is a greater openness to pursuing non-academic careers in these areas. It is possible there are things to learn from specific departments and programs who place Ph.D.s into diverse career areas that could be modeled by others. Certainly, providing diverse career pathways for Ph.D.s is critical as the "traditional" path of obtaining faculty positions becomes less available in many fields. 

Among the ~108,000 respondents to the NSF SDR 2019 survey who reported being employed at educational institutions in the US, 44.5% were tenured or on the tenure track <10 years since receiving their doctorate degree. This percentage jumped to 69.1% in those 10+ years from degree award. However, there were noticeable differences by degree field in the percentage of Ph.D.s employed at educational institutions who were tenured faculty or on the tenure-track <10 years from their Ph.D. with ~25% in this category among the biological, agricultural, and environmental life sciences to 60%+ for computer and information sciences and social sciences Ph.D.s. 

One might speculate that looking at these data for those <10 years from their Ph.D. points to a potential bottleneck to obtaining faculty positions among certain fields. Also, the length of postdoctoral positions and/or use of more contingent positions (lecturer, research associate) in educational institutions could be higher in some fields than others. The SDR data can offer some insights as the proportion of postdocs who are 5+ years from obtaining their Ph.D. is higher in the life sciences fields which also had the lowest proportion of Ph.D.s employed at educational institutions in tenured or tenure-track faculty positions (plotted in green in the graph below and above, respectively). 

While certainly the life sciences have the highest percentage of Ph.D.s employed as postdocs 5+ years from their Ph.D. and the lowest percentage of those <10 years from Ph.D. in tenure-track or tenured faculty roles, there is not perfect correspondence between lengthy postdocs and percentage of early-career Ph.D.s employed as tenured or tenure-track faculty. This could be for a myriad of reasons as the SDR data is not perfect. Remember, it only surveys individuals who earned their Ph.D.s in the United States. Thus, fields where a high percentage of workers obtain their Ph.D. outside the US are going to have less respondent representation in this survey.

For instance, we know that many postdoctoral scholars in the United States are international, who either obtained their Ph.D. in the US and continued in postdoctoral training via various visa types or who received their Ph.D. outside the US before doing a postdoc in the US. The 2019 SDR shows that ~54% of US Ph.D.s employed as postdocs are US citizens and other data from NSF shows ~49% of postdocs in the US were born oversees. In some fields including computer science and engineering, NSF estimates 55-60% of Ph.D.s working in those areas in the United States are foreign-born. Thus, the various employment trends shared so far can be affected by various limitations to employment for those individuals requiring visa sponsorship by their employer, the frequency of which may differ by Ph.D. field and the proportion of international students and scholars working in that area in the United States. I discussed some of the challenges around being an international scholar in the US (including visa restrictions) in an earlier series of blog posts.    

​Regardless of how international scholar dynamics may affect these data, it is clear from the 2019 SDR data that there are vast differences in the proportion of "early career" Ph.D.s in tenure-track or tenured faculty positions based on their degree field. ​

Much has been made of the decline in faculty positions available to Ph.D.s over recent years. The SDR data allows us to partially look at this trend by asking how the percentage of Ph.D. recipients employed at 4-year educational institutions has changed over the years. Here, I decided to look at the SDR data from 2010 and compare it to 2019.

Over the past 9 years the percentage of tenured faculty who are less than 10 years from the Ph.D. in most science fields has declined by 25-30%. The decline is less steep for tenure-track faculty in the life and physical/earth sciences. Furthermore, the proportion of engineering Ph.D.s <10 years from degree employed at educational institutions in tenure-track positions has actually increased from 2010 to 2019 based on the SDR data. Even in the engineering group, though, securing tenure by 10 years post degree has become less common, presumably as the need and/or length of postdoctoral positions have increased.  

The 2021 SDR data collection is currently underway and I will be very curious to see how these data look post-COVID. Will the percentages of early career Ph.D.s able to enter the faculty ranks fall even further? Only time will tell. 

As mentioned earlier, the prevalence of Ph.D. labor in the US who are supported on temporary visas is quite high. Many international students come to the US for their graduate training and seek employment in the country after finishing their degrees. The SDR data reports out median salaries for Ph.D. holders by citizenship status, which is plotted below by doctoral degree field. 

It is clear from these data that median salaries are lower, in aggregate, for temporary visa (J1, H1-B) holders in virtually all Ph.D. fields except mathematics & statistics AND social sciences. US permanent residents' median salaries also tend to be lower but not in all fields. In fact, in computer & information sciences and mathematics & statistics permanent residents earn slightly more than US citizens. 

It is difficult to speculate too much on these data but one potential reason for lower median salaries for temporary visa holders in particular could be the result of many of these individuals working at US universities where the visiting scholar (J1) visa category is commonly used when an individual is working as a postdoctoral scholar or some other contingent, non-tenure track position (research associate). When a temporary visa holder is employed by a company, however, they require H-1B sponsorship which is subject to a "prevailing wage" which should prevent these individuals' salaries being below "market" rate, at least in theory. The largest sponsors of H-1Bs in the US are typically companies working in the computer & information sciences or data analytics where Ph.D.s in the areas of computer science, math, and statistics would be in high demand. So, the increased salaries for temporary visa holders in those fields could be driven by who is employing the doctorate recipients (technology companies paying high wages).  

Beyond who employs Ph.D.s what work they do can drastically affect their level of compensation. As seen in the graph below, Ph.D. recipients whose primary work activity is teaching have lower median salaries than those in research & development (R&D) roles or focused more on professional services, administration, management, or sales. Clearly these data are also colored by who is employing Ph.D.s as teaching roles are almost entirely within universities whereas R&D roles could be at companies, universities, government agencies, or other employers. 

Note, however, that SDR data show the percentage of Ph.D.s whose primary work activity is teaching is ~10-15% of science and engineering Ph.D. recipients (see table, below). And there has been relatively little change in the percentages of Ph.D. recipients reporting their primary work role as teaching over the past few years. The general distribution of primary work roles for science and engineering Ph.D.s from 2017 to 2019 has remained relatively stable. And, as has been discussed in a previous post, the fact that greater than a third of science and engineering Ph.D.s report their primary work role as falling in areas outside research and development or teaching emphasizes the fact that there are many positions in administration, communications, management, and more that fall outside of the main boxes of teaching and research available to Ph.D. holders. I will be interested to see whether these distributions of work roles shift post-pandemic in the 2021 SDR data. Will there be less teaching roles? More R&D, especially in the life sciences? Or will the "something else" category continue to grow as Ph.D.s pursue more diverse career pathways?

The previous two salary graphs plot median salaries for all US Ph.D. recipients who completed the 2019 SDR. So, there are individuals in those data who are 20+ years from receiving their Ph.D.s and those who graduated only a few years ago. NSF also reports data by Ph.D. field filtered by years since doctorate which shows that the median salaries 5 years or less from being the Ph.D. awarded tend to hover around the $80,000 level though it is higher in some fields (most notably computer & information sciences). Median salaries are less different across Ph.D. science and engineering fields the further from the doctorate one looks. 

This final graph nicely illustrates the value Ph.D.s provide to their employers. One could speculate that as individuals with Ph.D. skills including critical thinking, problem solving, and knowledge synthesis also gain work experience post-Ph.D. employers value them more. Fifteen years from receiving their doctorate the median salary for all science & engineering Ph.D.s is $100,000+ and many are making well over that amount. Ph.D. training provides a valuable skillset when coupled with practical experience and knowledge of how to apply those skills through working with diverse employers. Perhaps training programs can do a better job of providing some of the practical skills valued by a variety of employers during graduate school to help aid Ph.D.s' transitions to employment after their degree? 

The NSF SDR data is an essential tool to help science and engineering graduate students, postdocs, and those who support them understand how the landscape of employment continues to evolve over time. Information on employment sectors and median salary data can also be helpful as recent Ph.D. recipients plot out the next step in their careers and understand their worth. 

There are certainly glaring issues that are evident in the NSF data as well. The fact that many Ph.D. recipients <10 years from their degree employed at educational institutions are not in tenure-track faculty or tenured faculty roles speaks to the erosion of the faculty career path for many. 

Furthermore, the proliferation of postdoctoral positions and other contingent roles is a problem. And while the number of those who received their Ph.D.s from US institutions officially employed in extended postdoctoral positions (5+ years post-Ph.D.) may be diminishing, we have less data on how many of these individuals have been captured by other job titles (such as research associate) when they "age-out" of the postdoc which may similarly lack pathways to permanent, well-compensated employment.

Certainly there are many unanswered questions in understanding the evolution of the Ph.D. workforce but NSF data provides critical insights which, when collected over time, allows for us to begin to observe changes in various employment metrics.

​I encourage you to explore the data for yourself at the links below.

As we approach the fall, faculty job ads will begin to be posted. How can you prepare yourself for a competitive job market? I have 5 pieces of advice from my own experience applying for faculty positions and work supporting postdocs in their job search. 

1. Self-assess & consider fit
2. Gather intel & information
3. Start early preparing application materials
4. Cultivate peer support
5. Consider diverse career options

Self-assess & consider fit
Before any career search process, you should spend time self-reflecting on your skills, interests, and values. Consider what types of roles and employers would best fit where you are in your life. Note that there are many roles in higher education beyond tenure-track faculty and there are many colleges & universities to work at beyond name-brand or research-intensive institutions.

For more see:

• Career Exploration 101
• Academic Career Readiness Assessment (ACRA) from UCSF
  • More on ACRA methodology 

Gather intel & information
What does it take to successfully land a tenure-track faculty position? While the ACRA referenced above assessed the faculty search committee's perspective on hiring new assistant professors, other work that I have collaborated on seeks to present data from the applicant's side.

See A survey-based analysis of the academic job market for more info on our applicant survey findings and variables that were associated with applicants successfully landing faculty job offers as part of the 2018-2019 job market cycle. 

Updated 6/29/2022:
If you were on the assistant professor job market in 2020-2021 and/or 2021-2022 (regardless of the outcome - whether you received a job offer or ultimately accepted one), we would love for you to complete our surveys as we seek to better understand how a variety of variables influence one's ability to land a faculty job offer. Find links to the surveys and learn more about our faculty job market collaboration team, below.

Arguably, the research statement is a good place to start for many STEM trainees as you probably have a good sense of what you have accomplished in your research and why it matters. However, a research statement needs to be forward looking and focused on what you plan to research as a new assistant professor. As such, you should think critically about where your field is going and how your training, skills, and interests can best be leveraged into a research program that is value-add to the department and institution you are joining. And if you are targeting research-intensive institutions, you must also seriously consider how your research will be funded and fit into funding priorities of key federal agencies and foundations.

• More on How to Construct a Compelling Research Statement

It's important to really think about your vision for your research first before tailoring your materials to a particular institution. I believe you shouldn't try to present the research you think a hiring committee will be excited about but rather research YOU are excited about. You want to be hired based on who you are and what you aspire to be and be OK with the realization that many hiring committees won't see a fit between your interests and theirs. That is OK because fit is so critical in any job search and you want to find it so that you can thrive in your new role and be your true self. 

For additional tips and advice on other faculty application documents, see:

• How to Write a Better Academic Cover Letter
• Tips on Writing an Effective Teaching Statement
• Diversity statement resources from UCLA & UCSD

And see the job search resources section of my website for documents I submitted on my own faculty job search, including a research statement for a research-intensive and primarily undergraduate-serving institution. 

Cultivate peer support
If you are currently a postdoctoral researcher interested in a faculty career, you should join Future PI Slack NOW. This community of nearly 4,000 postdocs is AMAZING and a great place to solicit advice and feedback from your peers. There is a channel in the Slack group devoted exclusively to reviewing one another's job application materials (#jobapp_reviewers) that you can join to solicit and provide feedback. It is a truly great way to see a variety of examples of faculty application materials and get additional feedback on your own documents from individuals who will often be outside your discipline. This type of feedback is critical as asking your current supervisor or lab-mates for feedback may not give you a sense of how someone outside your research domain views your documents...and faculty search committees will almost certainly be composed of a diverse set of individuals lacking a deep understanding of your particular area of research focus. So, getting broad-based feedback allows you to refine your documents to convey your points effectively to a wider audience.

You can also form your own peer support group at your institution or amongst others in your network. Find trainees also planning to go on the job market soon to meet with regularly to discuss your experiences and share your approach to the job search (where did you find job postings?, how are you handling requesting letters of reference?, etc...). Ideally, having this group be composed of individuals going on the job market this year and those planning to go on the job market next year is a great way for trainees a year out from applying to learn from those going through the process this year. As the hiring cycle progresses, your group can think learn from those who have had interviews with search committees about their experiences and serve as a space for feedback on the job talk and chalk talk (and see this presentation or this one on YouTube for more) your group members plan to deliver at institutions who have invited them for a final round interview. Through this experience, you'll gain a wider perspective on navigating a job search as you learn and support one another. 

Consider diverse career options
While a faculty role may be what you aspire to, it is always good to have options for the next step in your career. Consider using self-assessment tools such as myIDP, ImaginePhD, or ChemIDP to learn about the breadth of career paths that could fit your skills, interests, and values. Make plans to develop the skills and networks needed to pivot into careers beyond the tenure-track. There are so many amazing careers out there to pursue and remaining open to those possibilities will help you realize that there is no one path to "success" when it comes to your career. 

See my recent Standing Out from the Pack presentation for a summary of some of the resources and tools mentioned above as well as advice on navigating the faculty job search. 

We've also curated a variety of resources and perspectives on the academic job search and how to craft compelling application materials on the NC State Graduate School's ImPACKful blog under our "Academic PACKways" category. 

I also highly recommend this Faculty Corner YouTube playlist from Michigan Medicine's Office of Graduate & Postdoctoral Studies to learn more about the faculty job search and hear from faculty on how they manage their work & responsibilities.